Citrion measures who the AI assistants actually cite.
Every EZ Web AI visibility audit is produced by Citrion, the measurement instrument we built and run ourselves. This page says what it records, what comes out the other end, and what it is not.
Real answers, in real sessions, on five assistants.
Citrion asks the questions your buyers ask inside ChatGPT, Perplexity, Claude, Grok, and Copilot, and keeps the whole answer: the text, the sites it links to, and the search queries the assistant wrote for itself before answering. Nothing on the observed side is reconstructed from an API after the fact.
- Fresh sessions, our own accounts. Each run starts in a new chat with memory off, so nothing an assistant remembers about us can shape the answer.
- Fetched versus cited. The pages an assistant read are logged separately from the pages it credited, because being read and passed over is a different problem from being invisible.
- The queries behind the answer. An assistant searches its own phrasing, often including brand names the buyer never typed. Those phrasings are what your pages have to rank for.
- Crawler access. Citrion checks whether each assistant's crawler can read your site at all, and confirms a refusal against the assistant's own behavior before calling it a block.
- What it says about you. Every answer that names your business is read for how it describes you: favorable, neutral, mixed, or negative, with the sentence copied from the answer. Leads check a brand with an assistant before they sign, and this is what they hear.
Google AI Overviews are measured where we can observe them and modeled where we cannot. Every Google number in a report says which, and a modeled number is never presented as an observation.
One score, and the names behind it.
The share of recorded answers in which your business appeared, by name or by link. Computed from the answers, never estimated.
A leaderboard of the sites the assistants sent your buyers to instead, with the share each one holds. Usually one site holds a surprising amount of it.
Every answer that named you, read for favorability, with the drawbacks quoted word for word. The sentence an assistant attaches to your name is the one a lead reads before signing.
What to open, publish, and earn, in order, starting with anything that keeps the assistants from reading your site today.
Every number carries an evidence tier, published at how we measure AI visibility. Where a sample is too small to support a number, the report says so instead of printing one. Want to read one before you ask for yours? We ran Citrion on ezweb.dev and published the whole report, score and all.
An instrument, not a subscription.
Citrion is not a dashboard you log into, and we do not sell it on its own. It exists to produce the audit and to re-measure the same questions every month for clients on a plan, so the score moves against the same yardstick it started on.
It does not guarantee a citation. No instrument can, and attribution from an AI answer to a signed customer is still unsolved. Citrion measures visibility, which is the input you can actually change.
Platforms such as Profound and Evertune measure well and hand you a queue of recommendations. Citrion's output goes to the people who then do the work: the rewrites, the technical fixes, and the new pages, each reviewed by an expert before it ships. The protocol is public and the tooling stays private, so anyone can follow the recipe by hand and hold any other tool to the same standard.